Abstract
Due to the constantly growing interest in alternative investments, the art market has become the subject of numerous studies. By publishing sales data, many services and auction houses provide a foundation for further research on the latest trends. Determining the definition of the artistic value or formalisation of appraisal may be considered quite complex. Statistical analysis, econometric methods or data mining techniques could pave the way towards better understanding of the mechanisms occurring on the art market. The goal of this paper is to identify, describe and compare solutions (and related challenges) that help to analyse, make decisions and define state of the art in the context of the intersection of econometrics on art markets and computer science. This work is also a starting point for further research.
Introduction
The art market has historically been characterized by information asymmetry, opaque private transactions, and highly subjective valuations. Unlike traditional financial markets where assets are homogeneous, art is inherently non-homogeneous—no two pieces are identical. This uniqueness complicates the creation of reliable price indices. However, as art increasingly functions as a serious alternative investment—with some segments rivaling S&P 500 returns—there is a pressing need for objective, data-driven frameworks to measure market trends, evaluate portfolio diversification, and predict artist trajectories.
Analytical Methods for Art Valuation
Quantifying the value of art requires moving beyond expert opinion toward formal econometric and computational models. The paper identifies several key methodologies for constructing art market indices, ranging from simple averages to complex regressions.
The primary econometric approaches include:
- Repeat-sales Regression (RSR): This method tracks price changes of the same object sold multiple times. While highly accurate for the specific items tracked, it suffers from "data poverty," as only a small fraction (roughly 10%) of auction lots are repeat sales.
- Hedonic Regression (HR): Often considered the "gold standard" for non-homogeneous goods, HR decomposes a painting into its constituent characteristics—such as artist, size, medium, and historical "boom periods"—to determine the marginal contribution of each feature to the final price.
Beyond traditional econometrics, computer science techniques are being integrated:
- Data Mining & NLP: Researchers use N-gram analysis on large corpora (like Google Books) to correlate the "art investment lexicon" with actual market growth.
- Machine Learning: Emerging platforms utilize algorithms to forecast the future value of artists based on social mapping, museum presence, and exhibition history.
- Linked Open Data (LOD): Semantic Web standards are being explored to unify disparate data sources—from Wikipedia to auction results—into a single, structured model.
The Data Landscape
The shift toward data-oriented analysis is supported by several massive commercial and research databases:
- ArtPrice & Artnet: These provide millions of auction records dating back decades, covering hundreds of thousands of artists.
- Mei Moses Fine Art Index: A specialized dataset focused on repeat-sale lots, providing a long-term view of art as an asset class.
- Blouin Art Sales Index: Another significant repository of auction data.
Key Insights and Market Dynamics
Data-driven analysis has revealed several measurable phenomena in the art market:
- The "Death Effect": Econometric models confirm a statistically significant price jump following an artist's death.
- The RSR vs. HR Trade-off: While RSR is precise, it ignores the vast majority of the market. HR utilizes more data but must account for "selection bias," as auction records typically only reflect successful sales.
- Competitive Returns: Between 2003 and 2013, post-war and contemporary art delivered compound annual returns of 10.5% to 14.9%, making it a competitive alternative to traditional equities.
- Illiquidity: Data from the Mei Moses index shows an average 22-year cycle between the first and second sale of high-end artworks, highlighting the asset's extreme illiquidity.
Limitations and Future Outlook
Despite advancements, several challenges remain. Selection bias is a major hurdle; "bought-in" lots (items that fail to sell) are often excluded from public data, potentially inflating market indices. Furthermore, modern machine learning "signals" often operate as "black boxes," creating new forms of information asymmetry.
The future of art market analysis lies in standardization and automation. The adoption of formal ontologies and the Semantic Web will allow for a more structured set of characteristics beyond simple physical dimensions. Additionally, automating the creation of "comparable sets" will make price indices more robust and less dependent on manual expert intervention, paving the way for a truly transparent and data-driven art market.
Summary updated